[llvm] r298341 - [SystemZ] Don't drop MO flags in foldMemoryOperandImpl()
Jonas Paulsson via llvm-commits
llvm-commits at lists.llvm.org
Mon Mar 20 22:49:41 PDT 2017
Author: jonpa
Date: Tue Mar 21 00:49:40 2017
New Revision: 298341
URL: http://llvm.org/viewvc/llvm-project?rev=298341&view=rev
Log:
[SystemZ] Don't drop MO flags in foldMemoryOperandImpl()
The def operand of the new LG/LD should have the old def operands
flags and subreg index.
New test: test/CodeGen/SystemZ/fold-memory-op-impl.ll
Review: Ulrich Weigand
Added:
llvm/trunk/test/CodeGen/SystemZ/fold-memory-op-impl.ll
Modified:
llvm/trunk/lib/Target/SystemZ/SystemZInstrInfo.cpp
Modified: llvm/trunk/lib/Target/SystemZ/SystemZInstrInfo.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/SystemZ/SystemZInstrInfo.cpp?rev=298341&r1=298340&r2=298341&view=diff
==============================================================================
--- llvm/trunk/lib/Target/SystemZ/SystemZInstrInfo.cpp (original)
+++ llvm/trunk/lib/Target/SystemZ/SystemZInstrInfo.cpp Tue Mar 21 00:49:40 2017
@@ -1140,12 +1140,12 @@ MachineInstr *SystemZInstrInfo::foldMemo
// destination register instead.
if (OpNum == 1) {
unsigned LoadOpcode = Op0IsGPR ? SystemZ::LG : SystemZ::LD;
- unsigned Dest = MI.getOperand(0).getReg();
return BuildMI(*InsertPt->getParent(), InsertPt, MI.getDebugLoc(),
- get(LoadOpcode), Dest)
- .addFrameIndex(FrameIndex)
- .addImm(0)
- .addReg(0);
+ get(LoadOpcode))
+ .add(MI.getOperand(0))
+ .addFrameIndex(FrameIndex)
+ .addImm(0)
+ .addReg(0);
}
}
Added: llvm/trunk/test/CodeGen/SystemZ/fold-memory-op-impl.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/SystemZ/fold-memory-op-impl.ll?rev=298341&view=auto
==============================================================================
--- llvm/trunk/test/CodeGen/SystemZ/fold-memory-op-impl.ll (added)
+++ llvm/trunk/test/CodeGen/SystemZ/fold-memory-op-impl.ll Tue Mar 21 00:49:40 2017
@@ -0,0 +1,129 @@
+; RUN: llc < %s -mtriple=s390x-linux-gnu -mcpu=z13 -verify-machineinstrs | FileCheck %s
+;
+; Test that foldMemoryOperandImpl() doesn't drop subreg / read-undef flags.
+
+
+define void @fun_llvm_stress_reduced(i8*, i32*, i64*, i64, i8) {
+; CHECK: .text
+BB:
+ %A4 = alloca <4 x i64>
+ %A1 = alloca <8 x i1>
+ %E6 = extractelement <4 x i1> undef, i32 3
+ %L23 = load i8, i8* %0
+ %B27 = fmul double 0x59A989483BA7E0C6, undef
+ %L30 = load i16, i16* undef
+ store i16 -11933, i16* undef
+ %L46 = load i16, i16* undef
+ %L61 = load i16, i16* undef
+ %Sl74 = select i1 undef, i1 undef, i1 true
+ br label %CF846
+
+CF846: ; preds = %CF877, %BB
+ %I86 = insertelement <4 x i1> undef, i1 undef, i32 0
+ %Cmp89 = icmp ne i64 undef, %3
+ %L90 = load i16, i16* undef
+ %Shuff92 = shufflevector <4 x i16> zeroinitializer, <4 x i16> zeroinitializer, <4 x i32> <i32 0, i32 2, i32 undef, i32 6>
+ br label %CF877
+
+CF877: ; preds = %CF846
+ store i16 %L61, i16* undef
+ %Cmp110 = icmp eq i16 %L61, undef
+ br i1 %Cmp110, label %CF846, label %CF862
+
+CF862: ; preds = %CF877
+ %I114 = insertelement <4 x i64> zeroinitializer, i64 0, i32 0
+ %B115 = shl <4 x i64> zeroinitializer, %I114
+ %Sl124 = select i1 true, <8 x i1>* %A1, <8 x i1>* %A1
+ %B130 = frem double %B27, 0x59A989483BA7E0C6
+ %E143 = extractelement <4 x i64> %B115, i32 1
+ %Sl148 = select i1 %Cmp89, <1 x i32> undef, <1 x i32> zeroinitializer
+ br label %CF855
+
+CF855: ; preds = %CF855, %CF862
+ %Sl171 = select i1 %Sl74, i1 %E6, i1 undef
+ br i1 %Sl171, label %CF855, label %CF874
+
+CF874: ; preds = %CF855
+ %PC186 = bitcast i32* %1 to i16*
+ %L196 = load i16, i16* undef
+ %B207 = or i8 %4, %L23
+ %L211 = load <8 x i1>, <8 x i1>* %Sl124
+ %B215 = fdiv double 0x8421A9C0D21F6D3E, %B130
+ %L218 = load i16, i16* %PC186
+ %Sl223 = select i1 %Sl171, <4 x i1> %I86, <4 x i1> undef
+ br label %CF826
+
+CF826: ; preds = %CF866, %CF910, %CF874
+ %B245 = ashr i16 -11933, %L46
+ br label %CF910
+
+CF910: ; preds = %CF826
+ %L257 = load i8, i8* %0
+ %BC262 = bitcast i64 %E143 to double
+ store i16 %L196, i16* %PC186
+ %E266 = extractelement <4 x i16> %Shuff92, i32 0
+ %Sl271 = select i1 %Cmp89, i1 %Cmp89, i1 %Cmp110
+ br i1 %Sl271, label %CF826, label %CF866
+
+CF866: ; preds = %CF910
+ store i64 %E143, i64* %2
+ %I276 = insertelement <4 x double> undef, double %BC262, i32 3
+ %L281 = load <8 x i1>, <8 x i1>* %Sl124
+ %E282 = extractelement <4 x i1> zeroinitializer, i32 2
+ br i1 %E282, label %CF826, label %CF848
+
+CF848: ; preds = %CF866
+ %Cmp288 = fcmp olt <4 x double> undef, %I276
+ %FC294 = fptosi double undef to i16
+ %Cmp296 = icmp ule i16 %FC294, %B245
+ store i16 %L218, i16* undef
+ store i8 %L23, i8* %0
+ %E320 = extractelement <4 x i1> %Sl223, i32 1
+ %PC337 = bitcast <8 x i1>* %Sl124 to i1*
+ %Cmp345 = icmp uge <1 x i32> undef, %Sl148
+ store i16 %L196, i16* %PC186
+ br label %CF893
+
+CF893: ; preds = %CF893, %CF848
+ %Cmp361 = fcmp uge float undef, undef
+ br i1 %Cmp361, label %CF893, label %CF906
+
+CF906: ; preds = %CF893
+ store i16 -11933, i16* undef
+ %Shuff379 = shufflevector <1 x i1> undef, <1 x i1> %Cmp345, <1 x i32> <i32 1>
+ br label %CF850
+
+CF850: ; preds = %CF850, %CF906
+ br i1 undef, label %CF850, label %CF925
+
+CF925: ; preds = %CF850
+ store i16 %E266, i16* %PC186
+ %Cmp413 = icmp ugt i8 %L257, undef
+ store i16 %L30, i16* %PC186
+ %Sl420 = select i1 %Sl171, <8 x i1> undef, <8 x i1> %L281
+ store i16 %L90, i16* undef
+ %FC469 = uitofp i1 %Cmp296 to float
+ store i1 %Cmp413, i1* %PC337
+ br label %CF833
+
+CF833: ; preds = %CF833, %CF925
+ store i8 %B207, i8* %0
+ %E509 = extractelement <8 x i1> %L211, i32 7
+ br i1 %E509, label %CF833, label %CF882
+
+CF882: ; preds = %CF833
+ store i1 %Sl271, i1* %PC337
+ br label %CF852
+
+CF852: ; preds = %CF896, %CF882
+ store i1 %Sl74, i1* %PC337
+ br label %CF896
+
+CF896: ; preds = %CF852
+ %E576 = extractelement <4 x i1> %Cmp288, i32 3
+ br i1 %E576, label %CF852, label %CF890
+
+CF890: ; preds = %CF896
+ %Sl581 = select i1 undef, float undef, float %FC469
+ unreachable
+}
More information about the llvm-commits
mailing list